1 /**********************************************************
2 * Copyright 2008-2009 VMware, Inc. All rights reserved.
3 *
4 * Permission is hereby granted, free of charge, to any person
5 * obtaining a copy of this software and associated documentation
6 * files (the "Software"), to deal in the Software without
7 * restriction, including without limitation the rights to use, copy,
8 * modify, merge, publish, distribute, sublicense, and/or sell copies
9 * of the Software, and to permit persons to whom the Software is
10 * furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice shall be
13 * included in all copies or substantial portions of the Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
16 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
17 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
18 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
19 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
20 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
21 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 *
24 **********************************************************/
25
26 #include "util/u_inlines.h"
27 #include "util/u_prim.h"
28 #include "util/u_upload_mgr.h"
29 #include "indices/u_indices.h"
30
31 #include "svga_cmd.h"
32 #include "svga_draw.h"
33 #include "svga_draw_private.h"
34 #include "svga_resource_buffer.h"
35 #include "svga_winsys.h"
36 #include "svga_context.h"
37 #include "svga_hw_reg.h"
38
39
40 /**
41 * Return a new index buffer which contains a translation of the original
42 * index buffer. An example of a translation is converting from QUAD
43 * primitives to TRIANGLE primitives. Each set of four indexes for a quad
44 * will be converted to six indices for two triangles.
45 *
46 * Before generating the new index buffer we'll check if the incoming
47 * buffer already has a translated buffer that can be re-used.
48 * This benefits demos like Cinebench R15 which has many
49 * glDrawElements(GL_QUADS) commands (we can't draw quads natively).
50 *
51 * \param offset offset in bytes to first index to translate in src buffer
52 * \param orig_prim original primitive type (like PIPE_PRIM_QUADS)
53 * \param gen_prim new/generated primitive type (like PIPE_PRIM_TRIANGLES)
54 * \param orig_nr number of indexes to translate in source buffer
55 * \param gen_nr number of indexes to write into new/dest buffer
56 * \param index_size bytes per index (2 or 4)
57 * \param translate the translation function from the u_translate module
58 * \param out_buf returns the new/translated index buffer
59 * \return error code to indicate success failure
60 */
61 static enum pipe_error
translate_indices(struct svga_hwtnl * hwtnl,const struct pipe_draw_info * info,const struct pipe_draw_start_count_bias * draw,enum pipe_prim_type gen_prim,unsigned orig_nr,unsigned gen_nr,unsigned gen_size,u_translate_func translate,struct pipe_resource ** out_buf,unsigned * out_offset)62 translate_indices(struct svga_hwtnl *hwtnl,
63 const struct pipe_draw_info *info,
64 const struct pipe_draw_start_count_bias *draw,
65 enum pipe_prim_type gen_prim,
66 unsigned orig_nr, unsigned gen_nr,
67 unsigned gen_size,
68 u_translate_func translate,
69 struct pipe_resource **out_buf,
70 unsigned *out_offset)
71 {
72 struct pipe_context *pipe = &hwtnl->svga->pipe;
73 struct svga_screen *screen = svga_screen(pipe->screen);
74 struct svga_buffer *src_sbuf = NULL;
75 struct pipe_transfer *src_transfer = NULL;
76 struct pipe_transfer *dst_transfer = NULL;
77 const unsigned size = gen_size * gen_nr;
78 const unsigned offset = draw->start * info->index_size;
79 const void *src_map = NULL;
80 struct pipe_resource *dst = NULL;
81 void *dst_map = NULL;
82
83 assert(gen_size == 2 || gen_size == 4);
84 if (!info->has_user_indices)
85 src_sbuf = svga_buffer(info->index.resource);
86
87 /* If the draw_info provides us with a buffer rather than a
88 * user pointer, Check to see if we've already translated that buffer
89 */
90 if (src_sbuf && !screen->debug.no_cache_index_buffers) {
91 /* Check if we already have a translated index buffer */
92 if (src_sbuf->translated_indices.buffer &&
93 src_sbuf->translated_indices.orig_prim == info->mode &&
94 src_sbuf->translated_indices.new_prim == gen_prim &&
95 src_sbuf->translated_indices.offset == offset &&
96 src_sbuf->translated_indices.count == orig_nr &&
97 src_sbuf->translated_indices.index_size == gen_size) {
98 pipe_resource_reference(out_buf, src_sbuf->translated_indices.buffer);
99 return PIPE_OK;
100 }
101 }
102
103 /* Need to trim vertex count to make sure we don't write too much data
104 * to the dst buffer in the translate() call.
105 */
106 u_trim_pipe_prim(gen_prim, &gen_nr);
107
108 if (src_sbuf) {
109 /* If we have a source buffer, create a destination buffer in the
110 * hope that we can reuse the translated data later. If not,
111 * we'd probably be better off using the upload buffer.
112 */
113 dst = pipe_buffer_create(pipe->screen,
114 PIPE_BIND_INDEX_BUFFER, PIPE_USAGE_IMMUTABLE,
115 size);
116 if (!dst)
117 goto fail;
118
119 dst_map = pipe_buffer_map(pipe, dst, PIPE_MAP_WRITE, &dst_transfer);
120 if (!dst_map)
121 goto fail;
122
123 *out_offset = 0;
124 src_map = pipe_buffer_map(pipe, info->index.resource,
125 PIPE_MAP_READ |
126 PIPE_MAP_UNSYNCHRONIZED,
127 &src_transfer);
128 if (!src_map)
129 goto fail;
130 } else {
131 /* Allocate upload buffer space. Align to the index size. */
132 u_upload_alloc(pipe->stream_uploader, 0, size, gen_size,
133 out_offset, &dst, &dst_map);
134 if (!dst)
135 goto fail;
136
137 src_map = info->index.user;
138 }
139
140 translate((const char *) src_map + offset, 0, 0, gen_nr, 0, dst_map);
141
142 if (src_transfer)
143 pipe_buffer_unmap(pipe, src_transfer);
144
145 if (dst_transfer)
146 pipe_buffer_unmap(pipe, dst_transfer);
147 else
148 u_upload_unmap(pipe->stream_uploader);
149
150 *out_buf = dst;
151
152 if (src_sbuf && !screen->debug.no_cache_index_buffers) {
153 /* Save the new, translated index buffer in the hope we can use it
154 * again in the future.
155 */
156 pipe_resource_reference(&src_sbuf->translated_indices.buffer, dst);
157 src_sbuf->translated_indices.orig_prim = info->mode;
158 src_sbuf->translated_indices.new_prim = gen_prim;
159 src_sbuf->translated_indices.offset = offset;
160 src_sbuf->translated_indices.count = orig_nr;
161 src_sbuf->translated_indices.index_size = gen_size;
162 }
163
164 return PIPE_OK;
165
166 fail:
167 if (src_transfer)
168 pipe_buffer_unmap(pipe, src_transfer);
169
170 if (dst_transfer)
171 pipe_buffer_unmap(pipe, dst_transfer);
172 else if (dst_map)
173 u_upload_unmap(pipe->stream_uploader);
174
175 if (dst)
176 pipe_resource_reference(&dst, NULL);
177
178 return PIPE_ERROR_OUT_OF_MEMORY;
179 }
180
181
182 enum pipe_error
svga_hwtnl_simple_draw_range_elements(struct svga_hwtnl * hwtnl,struct pipe_resource * index_buffer,unsigned index_size,int index_bias,unsigned min_index,unsigned max_index,enum pipe_prim_type prim,unsigned start,unsigned count,unsigned start_instance,unsigned instance_count,ubyte vertices_per_patch)183 svga_hwtnl_simple_draw_range_elements(struct svga_hwtnl *hwtnl,
184 struct pipe_resource *index_buffer,
185 unsigned index_size, int index_bias,
186 unsigned min_index, unsigned max_index,
187 enum pipe_prim_type prim, unsigned start,
188 unsigned count,
189 unsigned start_instance,
190 unsigned instance_count,
191 ubyte vertices_per_patch)
192 {
193 SVGA3dPrimitiveRange range;
194 unsigned hw_prim;
195 unsigned hw_count;
196 unsigned index_offset = start * index_size;
197
198 hw_prim = svga_translate_prim(prim, count, &hw_count, vertices_per_patch);
199 if (hw_count == 0)
200 return PIPE_OK; /* nothing to draw */
201
202 range.primType = hw_prim;
203 range.primitiveCount = hw_count;
204 range.indexArray.offset = index_offset;
205 range.indexArray.stride = index_size;
206 range.indexWidth = index_size;
207 range.indexBias = index_bias;
208
209 return svga_hwtnl_prim(hwtnl, &range, count,
210 min_index, max_index, index_buffer,
211 start_instance, instance_count,
212 NULL, NULL);
213 }
214
215
216 enum pipe_error
svga_hwtnl_draw_range_elements(struct svga_hwtnl * hwtnl,const struct pipe_draw_info * info,const struct pipe_draw_start_count_bias * draw,unsigned count)217 svga_hwtnl_draw_range_elements(struct svga_hwtnl *hwtnl,
218 const struct pipe_draw_info *info,
219 const struct pipe_draw_start_count_bias *draw,
220 unsigned count)
221 {
222 struct pipe_context *pipe = &hwtnl->svga->pipe;
223 enum pipe_prim_type gen_prim;
224 unsigned gen_size, gen_nr;
225 enum indices_mode gen_type;
226 u_translate_func gen_func;
227 enum pipe_error ret = PIPE_OK;
228
229 SVGA_STATS_TIME_PUSH(svga_sws(hwtnl->svga),
230 SVGA_STATS_TIME_HWTNLDRAWELEMENTS);
231
232 if (svga_need_unfilled_fallback(hwtnl, info->mode)) {
233 gen_type = u_unfilled_translator(info->mode,
234 info->index_size,
235 count,
236 hwtnl->api_fillmode,
237 &gen_prim,
238 &gen_size, &gen_nr, &gen_func);
239 }
240 else {
241 unsigned hw_pv;
242
243 /* There is no geometry ordering with PATCH, so no need to
244 * consider provoking vertex mode for the translation.
245 * So use the same api_pv as the hw_pv.
246 */
247 hw_pv = info->mode == PIPE_PRIM_PATCHES ? hwtnl->api_pv :
248 hwtnl->hw_pv;
249 gen_type = u_index_translator(svga_hw_prims,
250 info->mode,
251 info->index_size,
252 count,
253 hwtnl->api_pv,
254 hw_pv,
255 PR_DISABLE,
256 &gen_prim, &gen_size, &gen_nr, &gen_func);
257 }
258
259 if ((gen_type == U_TRANSLATE_MEMCPY) && (info->index_size == gen_size)) {
260 /* No need for translation, just pass through to hardware:
261 */
262 unsigned start_offset = draw->start * info->index_size;
263 struct pipe_resource *index_buffer = NULL;
264 unsigned index_offset;
265
266 if (info->has_user_indices) {
267 u_upload_data(pipe->stream_uploader, 0, count * info->index_size,
268 info->index_size, (char *) info->index.user + start_offset,
269 &index_offset, &index_buffer);
270 u_upload_unmap(pipe->stream_uploader);
271 index_offset /= info->index_size;
272 } else {
273 pipe_resource_reference(&index_buffer, info->index.resource);
274 index_offset = draw->start;
275 }
276
277 assert(index_buffer != NULL);
278
279 ret = svga_hwtnl_simple_draw_range_elements(hwtnl, index_buffer,
280 info->index_size,
281 draw->index_bias,
282 info->index_bounds_valid ? info->min_index : 0,
283 info->index_bounds_valid ? info->max_index : ~0,
284 gen_prim, index_offset, count,
285 info->start_instance,
286 info->instance_count,
287 hwtnl->svga->patch_vertices);
288 pipe_resource_reference(&index_buffer, NULL);
289 }
290 else {
291 struct pipe_resource *gen_buf = NULL;
292 unsigned gen_offset = 0;
293
294 /* Need to allocate a new index buffer and run the translate
295 * func to populate it. Could potentially cache this translated
296 * index buffer with the original to avoid future
297 * re-translations. Not much point if we're just accelerating
298 * GL though, as index buffers are typically used only once
299 * there.
300 */
301 ret = translate_indices(hwtnl, info, draw, gen_prim,
302 count, gen_nr, gen_size,
303 gen_func, &gen_buf, &gen_offset);
304 if (ret == PIPE_OK) {
305 gen_offset /= gen_size;
306 ret = svga_hwtnl_simple_draw_range_elements(hwtnl,
307 gen_buf,
308 gen_size,
309 draw->index_bias,
310 info->index_bounds_valid ? info->min_index : 0,
311 info->index_bounds_valid ? info->max_index : ~0,
312 gen_prim, gen_offset,
313 gen_nr,
314 info->start_instance,
315 info->instance_count,
316 hwtnl->svga->patch_vertices);
317 }
318
319 if (gen_buf) {
320 pipe_resource_reference(&gen_buf, NULL);
321 }
322 }
323
324 SVGA_STATS_TIME_POP(svga_sws(hwtnl->svga));
325 return ret;
326 }
327